Dataset for detecting motorcyclists in pedestrian areas

Area of Science:

• Computer Science
• Artificial Intelligence
• Internet of Things (IoT)

Background:

• Road safety is a critical concern, particularly for vulnerable road users like pedestrians and motorcyclists.
• Existing object detection systems often require large, well-labeled datasets, which are time-consuming and expensive to create.
• The integration of IoT devices and artificial vision offers a novel approach to data acquisition for traffic analysis.

Purpose of the Study:

• To develop a semi-automated, scalable methodology for collecting and processing road imagery using IoT devices.
• To construct a multiclass object classifier for identifying pedestrians and motorcyclists.
• To create an efficient data pipeline for generating labeled datasets from public CCTV feeds.

Main Methods:

• Implementation of an Internet of Things (IoT) system using Python for continuous image acquisition from public CCTV cameras.
• Application of artificial vision techniques for automated image analysis and chronological data sorting.
• Development of two algorithms for dataset debugging: one for user-assisted labeling via Regions of Interest (ROI) and hotkeys, and another for verification of labeled data.

Main Results:

• Successful acquisition and processing of asynchronous image data from 80 CCTV cameras in Medellin, Colombia.
• Creation of a semi-automated pipeline for generating a debugged dataset suitable for training machine learning models.
• Demonstration of a scalable methodology for building multiclass object classifiers for road safety applications.

Conclusions:

• The proposed IoT-based methodology offers a scalable and efficient solution for creating object detection datasets from real-world traffic scenarios.
• The semi-automated approach significantly reduces the effort required for data labeling, accelerating the development of road safety AI.
• This work contributes to improving road safety by enabling the construction of more accurate multiclass object classifiers.